Demo prototype β€” not a real clinic

Let your clients book
24/7 on WhatsApp
β€” fully automated

Healthline is an AI assistant that handles appointment booking, rescheduling, cancellations, and customer Q&A automatically on WhatsApp β€” no extra staff needed.

✦ Interactive Demo β€” fictional clinic
πŸŽ™οΈ Voice works on WhatsApp β€” text only here πŸ“² Want the real WhatsApp demo? Contact us

What Healthline does

Everything handled. Automatically.

From the first "Hello" to a confirmed appointment β€” your clients get instant answers and seamless booking, around the clock.

πŸ“…

Instant Booking

Clients book in plain language. The AI finds the slot and confirms β€” no forms, no calls.

πŸ”„

Reschedule & Cancel

Clients manage their appointments with a message. Your policies are enforced automatically.

πŸ’¬

24/7 Q&A

Pricing, hours, services β€” the AI answers instantly from your knowledge base, day or night.

🌍

Timezone-Aware

Clients book in their own timezone. The AI converts and validates against your business hours.

πŸŽ™οΈ

Voice Messages

Clients can speak instead of type. The AI transcribes and responds β€” fully on WhatsApp.

πŸ”’

Secure by Design

Each client sees only their own data. Protected against manipulation and unauthorized access.

Any business that takes appointments

If your clients book time with you, Healthline can automate it β€” regardless of your industry.

πŸ₯ Clinics & Doctors
πŸ’† Spas & Wellness
πŸ’‡ Salons & Barbershops
πŸ‹οΈ Fitness & Coaches
βš–οΈ Law & Consulting
🏑 Real Estate
🐾 Vets & Pet Care
πŸ“Έ Photographers

How it works

Simple setup, immediate impact

Tailored to your business β€” we handle the configuration so you don't have to.

1

We learn your business

Tell us your services, hours, pricing, and policies. We configure the AI to know your business inside out.

2

We connect your WhatsApp

Your existing WhatsApp Business number gets connected to Healthline. Nothing changes for your clients.

3

Clients message as usual

They send a WhatsApp message. The AI handles it instantly β€” booking, answering, confirming β€” 24/7.

βš™οΈ For Technical Readers

How it's actually built

A production-grade AI system β€” not a chatbot wrapper. Here's what makes it work.

πŸ”€

LangGraph Workflow Engine

  • Conversation modelled as a directed graph
  • Each node has one job β€” no monolithic prompt
  • intent β†’ book / cancel / reschedule / Q&A β†’ confirm
  • List appointments β€” past, future, or both
  • Shared state carries context across every step
LangGraph LangChain FastAPI
πŸ› οΈ

LangChain Tool Calling

  • AI calls real tools mid-conversation, not just generates text
  • Google Calendar β€” availability, slot picking, event creation
  • Auto timezone conversion for every user
  • Capped at 5 tool calls per turn to control cost
Google Calendar API LangChain MCP
πŸ’¬

WhatsApp Webhook Integration

  • Real-time messages via WhatsApp Cloud API webhook
  • Webhook verification + duplicate filtering built-in
  • Text and voice in the same async pipeline
WhatsApp Cloud API FastAPI
πŸŽ™οΈ

Voice Message Support

  • Voice notes transcribed via Groq Whisper
  • Transcript handled exactly like a text message
  • Audio file deleted immediately after transcription
Groq Whisper WhatsApp
πŸ“š

Knowledge Base & RAG

  • Business info injected as context for Q&A
  • Fast, zero retrieval overhead for focused domains
  • Extensible to full vector RAG β€” no architecture changes needed
LangChain OpenAI Embeddings ChromaDB
πŸ’Ύ

Two-Layer Memory

  • Short-term: full conversation state in memory, per user
  • Long-term: client profiles in PostgreSQL β€” survives restarts
  • Sessions keyed by phone number β€” never mixed
LangGraph PostgreSQL Redis
πŸ”’

Authentication & Security

  • Auth via WhatsApp phone number β€” no password needed
  • Each user sees only their own data
  • Explicit confirmation before every sensitive action
  • Secrets in GCP Secret Manager β€” never in code
WhatsApp GCP Secret Manager
☁️

GCP Deployment (IaC)

  • Full infra as code β€” deployed in one command
  • Cloud Run β€” scales to zero when idle
  • Private Docker registry, zero hardcoded config
Google Cloud Run Terraform Docker

Dual-LLM split β€” reasoning vs. generation

Reasoning LLM
Claude Sonnet (Anthropic)
Structured output only β€” intent classification, appointment scope, slot selection, confirmation gating. Returns typed Pydantic schemas, never free text.
Generation LLM
GPT-4.1 mini (OpenAI)
All natural-language responses and tool calling. Handles the conversational layer and calendar/timezone/audio tool loops.